Check out the new USENIX Web site. next up previous
Next: Behavior of Uncoordinated TCP Up: Evaluation Previous: Evaluation

Network Topology

Figure 6: Simulation testbed in ns2.
\begin{figure}\centerline{\scalebox{0.8}{\epsfig{file=exp_network.eps}}}\end{figure}

Our simulation topology is pictured in Figure 6. A cluster of sending agents is labeled $S_1$ through $S_n$, with its local aggregation point labeled $AP_S$. A remote cluster of ACK (acknowledgement) agents is labeled $A_1$ through $A_n$, with its aggregation point labeled $AP_A$. $I_1$ and $I_2$ are intermediary nodes used to create a congested link, and $T_1$ and $T_2$ are used for traffic generation.

Propagation delay on links $AP_S$-$I_1$, $I_1$-$I_2$, and $I_2$-$AP_A$ is configured to be 4 msec, while it is only 1 msec on links $S_i$-$AP_S$ and $A_i$-$AP_A$. The link capacity for all links is 10 Mb/s, except for links $T_1$-$I_1$ and $T_2$-$I_2$ where link capacity is 100 Mb/s. This allows traffic generators to increase traffic over link $I_1$-$I_2$ to any desired level.

Trace data is collected as it is transmitted from $AP_S$ to $I_1$ since this allows us to observe sending rates before additional traffic on the link $I_1$-$I_2$ causes queuing delays, drops, or jitter not reflective of cluster endpoint sending rates.

TCP and C-TCP flows in this section use an infinitely large data source and send at the maximum rate allowed by their respective algorithms. Congestion periods are created by configuring $T_1$ and $T_2$ to generate constant bitrate traffic across the link $I_1$-$I_2$. In particular, a CBR agent sending at a constant 7.5-9.0 Mb/s from $T_1$ to $T_2$ competes with data traffic from $S_1$-$S_n$ over link $I_1$-$I_2$.


next up previous
Next: Behavior of Uncoordinated TCP Up: Evaluation Previous: Evaluation
David Ott 2002-04-16